CN-122021270-A - Asphalt pavement rut depth prediction and pavement structure design method capable of explaining PIML energization

CN122021270ACN 122021270 ACN122021270 ACN 122021270ACN-122021270-A

Abstract

The invention relates to the technical field of asphalt pavement rut depth prediction and pavement structure design, in particular to an asphalt pavement rut depth prediction and pavement structure design method capable of interpreting PIML energized, wherein an output variable of a mechanical experience model PHY model is used as physical priori knowledge to be integrated into a data-driven ML model training process, a physical information-driven machine learning prediction model is constructed, the mechanism reliability of the physical model and the strong nonlinear fitting capacity of the machine learning model are effectively integrated, the trained PHY-ML model is subjected to interpretative analysis by introducing an SHAP algorithm, the specific contribution degree and the influence direction of each characteristic parameter to rut depth prediction results are quantized and revealed, clear basis is provided for understanding key driving factors formed by rut, an improvement direction is provided for subsequent pavement structure design, and the technical problems that the prediction accuracy of the conventional rut depth prediction model is poor, and the prediction results are difficult to be converted into guiding basis of the structure design in the prior art are solved.

Inventors

LV HUIJIE
BAI JIAPENG
Hui Yingxin
LI BOTAN

Assignees

宁夏大学

Dates

Publication Date: 20260512
Application Date: 20260114

Claims (8)

1. A method for predicting rut depth of an asphalt pavement, the method comprising the steps of: s1, acquiring external factors and internal structure data of different pavement structures to construct a feature data set containing a plurality of features, and constructing a prediction target data set containing a plurality of prediction targets according to track depths of different pavement structures; S2, calculating correlation coefficients between each feature and each prediction target and between different features, and screening the features in the feature data set; S3, constructing a PHY model based on the screened characteristic data set, and outputting PHY variables; S4, taking PHY variables as physical priori knowledge to be integrated into the screened characteristic data set, training a PHY-ML model constructed based on the PHY model through a model training strategy, and optimizing model parameters by adopting Optuna super-parameter optimization algorithm and five-fold cross-validation method to obtain a trained PHY-ML model; S5, acquiring a characteristic data set of the asphalt pavement to be predicted, and inputting the trained PHY-ML model to obtain a rutting depth prediction result.
2. The method of claim 1, wherein the external factors include cumulative number of axle loads and average temperature of the pavement structure.
3. The method of claim 1, wherein the internal structural data comprises dynamic modulus of each layer of material and each layer thickness of the pavement structure.
4. The method according to claim 1, wherein the number of structural layers provided for the different pavement structures is the same, and the feature value of the corresponding structural layer is set to 0 if the corresponding structural layer does not exist for the different pavement structures.
5. The method of claim 1, wherein the pavement structure comprises a strong base thin-face semi-rigid substructure, a rigid substructure and a stress absorbing layer, a conventional asphalt pavement structure, a full thickness asphalt pavement structure, and a combination of asphalt concrete and semi-rigid substructure.
6. The method for predicting the rut depth of an asphalt pavement according to claim 1, the method is characterized in that the expression of the PHY model is as follows: In the above-mentioned method, the step of, Indicating the depth of the rut, Is the equivalent modulus of the asphalt layer material, Is the total thickness of the asphalt layer, Is the equivalent modulus of the non-asphalt layer material, Is the total thickness of the non-asphaltic layer, For the number of times the axle load is accumulated, In order to be able to achieve an average temperature, For the modulus of the material of each layer, The total number of the pavement structure layers is that the thickness of each layer is , For the total thickness of each pavement structure layer, In order to calibrate the coefficients of the light-emitting diode, Is a temperature coefficient of the silicon carbide material, As the coefficient of the axial load, Is the thickness coefficient of the asphalt layer, Is a non-asphalt layer thickness coefficient, Is the equivalent modulus coefficient of the asphalt layer, Is the equivalent modulus coefficient of the non-asphalt layer.
7. A method for constructing an interpretable PIML-energized bituminous pavement structure based on the trained PHY-ML model of any one of claims 1-6, comprising the steps of: performing interpretive analysis on the trained PHY-ML model by adopting an SHAP algorithm, and calculating SHAP values corresponding to all feature parameters in the feature data set fused with PHY variables; selecting a plurality of characteristic parameters with the largest mean value of SHAP values as typical characteristic parameters, and analyzing the influence of the typical characteristic parameters on the rut depth of the asphalt pavement; setting a target value of rut depth and a design parameter set serving as a characteristic data set thereof; inputting the design parameter set into the trained PHY-ML model to obtain a predicted rut depth; And selecting a plurality of typical characteristic parameters to carry out parameter adjustment according to the influence result of the typical characteristic parameters on the rut depth of the asphalt pavement until the predicted rut depth obtained after the adjusted design parameter set is input into the trained PHY-ML model again reaches the target value of the rut depth.
8. The asphalt pavement structure design method according to claim 7, wherein the calculation formula of the SHAP value is as follows: In the above-mentioned method, the step of, Is the characteristic parameter Is used to determine the SHAP value of (1), For a set of all of the characteristic parameters, To not include characteristic parameters Is a subset of the features of (a), To not include characteristic parameters Is provided for the set of all the characteristic parameters of (a), To use feature subsets And characteristic parameters The model output at the time of the prediction is made, To use only feature subsets And outputting a model when predicting the characteristic parameters.

Description

Asphalt pavement rut depth prediction and pavement structure design method capable of explaining PIML energization Technical Field The invention relates to the technical field of asphalt pavement track depth prediction and pavement structure design, in particular to an asphalt pavement track depth prediction and pavement structure design method capable of explaining PIML energization. Background Asphalt pavement ruts are typical structural diseases in the service period of roads, and management and prediction of the asphalt pavement ruts are always the research emphasis of pavement structural design and maintenance management. The rut hazard is essentially formed by longitudinal permanent deformation caused by accumulated irreversible plastic flow of asphalt mixture under the coupling action of repeated traffic load and complex environmental conditions, and the complexity of a deformation mechanism is further increased by multi-factor nonlinear interaction of different environmental conditions, pavement structure forms and the like. Along with the continuous rising of the proportion of heavy traffic and frequent extreme climate events, the realization of accurate prediction and active prevention and control of rut hazard gradually becomes a key link for formulating scientific maintenance strategies and prolonging the service life of roads. Accurate prediction of asphalt pavement rut depth is an important precondition for managing rut hazards. The traditional prediction method mainly comprises a mechanical-experience model and finite element simulation, wherein the mechanical-experience model is based on engineering experience and mechanical principle, pavement response is simplified into elasticity or linear viscoelasticity, a mathematical expression is constructed to describe rut development, calculation is simple and convenient but universality is limited, and the finite element simulation model is capable of finely simulating pavement behavior but is complex in calculation and sensitive in parameters. In contrast, machine learning algorithms, by virtue of their powerful nonlinear mapping capability and multivariate processing advantages, can effectively establish complex quantitative relationships between feature variables and target parameters. The effective and reasonable structural design is a core step of improving the rut hazard and ensuring the durability of the pavement. The traditional pavement structure design method mainly depends on engineering experience or parameter ranges recommended by design specifications, parameters are independently selected, the synergistic effect of design parameters and external environment changes on the parameters cannot be systematically considered, accurate description of complex nonlinear accumulated damage mechanisms among materials, environments and traffic loads is absent, and the finally designed pavement structure is difficult to meet normal engineering requirements. With the development of engineering technology and the rising of machine learning, the pavement design concept is gradually changed from traditional 'standard design' to 'optimal design based on prediction performance', so that obtaining high-precision performance prediction results and how to convert the high-precision performance prediction results into guiding basis of structural design become the key for improving the pavement structural design level. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a method for predicting the rut depth of an asphalt pavement and designing a pavement structure, which can explain PIML energization, and solves the technical problems that the existing rut depth prediction model is poor in prediction precision and the prediction result is difficult to be converted into a guiding basis of structural design in the prior art. In order to solve the technical problems, the invention provides the following technical scheme that the method for predicting the track depth of the asphalt pavement comprises the following steps: s1, acquiring external factors and internal structure data of different pavement structures to construct a feature data set containing a plurality of features, and constructing a prediction target data set containing a plurality of prediction targets according to track depths of different pavement structures; S2, calculating correlation coefficients between each feature and each prediction target and between different features, and screening the features in the feature data set; S3, constructing a PHY model based on the screened characteristic data set, and outputting PHY variables; S4, taking PHY variables as physical priori knowledge to be integrated into the screened characteristic data set, training a PHY-ML model constructed based on the PHY model through a model training strategy, and optimizing model parameters by adopting Optuna super-parameter optimization algorithm and five-fold cross-validation meth